why train loss will be nan? - Githubissues

CharlieDinh / pFedMe

Personalized Federated Learning with Moreau Envelopes (pFedMe) using Pytorch (NeurIPS 2020)

289 stars 89 forks source link

why train loss will be nan? #11

Closed GaoTiaoKang closed 3 years ago

GaoTiaoKang commented 3 years ago

why train loss will be nan?

CharlieDinh commented 3 years ago

Hi, the training loss is nan when the learning rate is large.